Full-text reading & foundational papers

BioSkepsis does not stop at the abstract. As it works through a question, the agent reads entire papers — methods, results, and discussion — and surfaces the foundational works most co-cited by your sources, so a brief rests on what a paper actually shows and the literature it is built on.

#Reading entire papers, not abstracts

Most literature tools reason over titles and abstracts because that is the only text they can retrieve at scale. An abstract is a marketing summary of a paper: it states the headline result but omits the sample size, the controls, the statistical caveats, and the conditions under which the finding does or does not hold. Reasoning from abstracts alone reliably overstates evidence.

Instead, when full text is available, the agent retrieves and reads the complete paper. It does not treat the document as one undifferentiated block of text — it works through the structured sections that carry the real evidence:

• Methods. The experimental design, model system, cohort, dose, controls, and analytical approach — the details that decide whether a result is causal, correlational, or preclinical.
• Results. The actual measurements, effect sizes, and statistics, rather than the author's one-line gloss of them in the abstract.
• Discussion. The interpretation, the boundary conditions, and — crucially — the author-stated limitations and open questions that abstracts routinely leave out.

#Why full text matters

Reading the whole paper changes the quality of every downstream step in a research run. Because the agent has the methods and results in front of it, it can:

• Verify claims against the source passage. A citation is only marked full-text-verified once the agent has matched the specific claim to the exact passage that supports it — something impossible when only an abstract is available.
• Type the evidence correctly. Whether a finding is causal, associational, synthesis, or preclinical is usually decided in the methods, not the abstract. Full-text reading is what lets the agent weigh a randomized trial differently from a cell-line experiment.
• Capture caveats and contradictions. Limitations, failed replications, and conflicting sub-results live in the results and discussion. Surfacing them is how the agent flags contradictions and evidence gaps instead of papering over them.

#Foundational papers via co-citation analysis

As the agent reads your sources, it also looks at what those sources cite. Foundational papers are the works your retrieved literature collectively builds on — the seminal studies, standard methods, and defining reviews that a field is quietly resting on. The agent finds them automatically using co-citation analysis across the life-science citation graph, so you never miss the bedrock references even when they predate the current results.

The core idea is simple: the more of your sources that cite the same earlier paper, the more foundational that earlier paper is to the topic. The agent walks the reference lists of your results, tallies how often each referenced work appears, and promotes the papers with the highest shared co-citation. It then refines that ranking with the quality of the citations, not just their count:

• Co-citation frequency. How many of your sources cite the same candidate — the primary signal of a shared intellectual foundation.
• Citation influence. Whether those citations are marked as influential, meaning the cited work meaningfully shaped the citing paper rather than appearing in a passing list of references.
• Methodology relevance. Whether the candidate is cited in a methods context, which flags papers that introduced a widely adopted technique or protocol.

Combining these signals surfaces papers that are both widely cited and meaningfully cited — the seminal discoveries, standard methodologies, and canonical reviews that define an area, rather than whatever simply happens to be old.

#How the agent uses them

Foundational papers are not a side report you have to go read separately. The agent treats them as candidate sources: the strongest foundational works can be folded into the evidence base for your question, read in full text like any other source, and cited in the brief. That gives an answer two layers of grounding at once — the latest research and the seminal work behind it — for a result that is both current and historically anchored.

In practice this is what lets a brief explain not just what a recent study found, but why the field frames the problem the way it does, which method everyone inherited, and where the original claim came from. For entering an unfamiliar area, the foundational set doubles as the essential reading list — the papers everyone in the field already knows.